Game AI with decision models

Jev beat Pokémon Red in 37h 40m over 16,150 decisions for about $1.65, picking every move from harness-listed legal options.

AI town simulation forked from a16z ai-town where Jev picks each agent's next legal move and an LLM handles dialogue and memory.

Twenty-questions game where Jev's confidence decides when to stop: it commits, hedges or refuses to guess.

Japanese quiz demo where Jev guesses which of AWS, Azure or Google Cloud a debranded service name belongs to, showing its probabilities.

Small demo collection testing Jev's spatial reasoning with maze single-step choices versus multi-step lookahead.

Browser Doom experiment where two WebAssembly engines play from the same start and Jev picks tactical macros from structured state.

Live demo where Jev plays Flappy Bird alone, answering flap or wait about twenty times a second with several requests in flight.

MoonBit client for Jev with a CLI and a Jev-versus-Jev gomoku match, plus a tool turning timed game logs into GIFs.

Game of Thrones roleplay where an LLM writes each scene and Jev labels location, beat, mood, danger and whether prose stayed in fiction.

Local Snake game where code computes facts for each legal move and a local Qwen3.5 picks one in a single Jev-style Choice pass.

WebGL card game Speed against a Jev-powered CPU, showing Jev's decision latency and accuracy against a rule validator in real time.

Chromium T-Rex Runner game where Jev chooses jump, duck or keep running for each obstacle, plus short or full jump.

Streamlit Tetris demo where Jev picks one collision-checked placement ID and Python verifies cells before locking.

3D Magic 8 Ball you shake and release, where Jev chooses among the 20 classic answers and shows a probability for each.

Desktop Snake where one Jev call per tick asks nine hazard and food questions for left, straight and right, and Python combines them.

Terminal Sudoku where Python owns the rules and Jev picks one move per turn, solving easy boards but struggling once it must guess.

Experimental controller where Jev picks NES inputs for Super Mario Bros. from structured emulator RAM and telemetry, not screenshots.

MoonBit and TypeScript Jev playground: gomoku, shell-command risk gate, ESLint plugin, a tiny Jev-condition language and measured reports.

Experimental Minecraft bot where each TypeSafe decision picks one primitive action, which completed a 338-block flag build in recorded runs.

Open family of System One-style models fine-tuned from Qwen3.5 and Gemma that return typed decision probabilities in one pass.

Independent MLX port of Laya for Apple Silicon: typed decisions in about 13 ms (7 ms multilingual) with no PyTorch or cloud API.

Apache-2.0 395M ModernBERT System One model serving calibrated Choice, Noul and Score locally as a drop-in /v1/systemone replacement.

Deploys DiffusionGemma-Jev behind a TypeSafe-compatible API on a Cloud Run GPU, with snake, dino and tetris demos.

Kotlin NES emulator whose agent plays Super Mario Bros. and Final Fantasy, with a local model choosing from the goals that apply each turn.

Gemma 3 4B option scorer that prefills context once and scores all options in one pass, on MLX or PyTorch, with a Doom demo.

Chinese local MLX demo that scores 2-5 candidate decisions with a small Qwen model and drives a Snake game by choosing among moves.

0.6B open Jev replica returning full option distributions with no decoding; one checkpoint plays ViZDoom, Maze and Snake.

Educational MLX project scoring typed questions about one image with Qwen3.5-0.8B on Apple Silicon, with game demos.

Reinforcement-learning lab where cached Jev judgments act as reward signals for DQN agents on four classic control and grid games.

Training library for a small one-pass model that scores a changing list of text options and returns a probability per option.